A Registration Method For Fluorescein Fundus Angiography Sequences And Its Applications

The fundus is the only tissue in the human body through which the blood vessels can be directly observed without invasive surgery.Fluorescein angiography(FFA)is the gold standard for the diagnosis of fundus diseases,and it is an important technique to obtain the fundus structure and to observe retinal blood circulation.Quantitative analysis of FFA images is the basis for follow-up observation and accurate evaluation of fundus diseases.However,during the sampling of FFA images,the patient inevitably moves the eyeball,resulting in changes in the position of the region of interest(such as retinal vessel branches and new blood vessels)in the FFA image sequence,which makes it difficult to quantitatively analyze the subsequent images and to accurately assess the disease in fundus.Traditional image registration methods have certain limitations in dealing with FFA image sequence,because the FFA image sequence is characterized by background fluorescence noise,large spatial position of blood vessels and different perfusion state at different time points.In order to solve the above problems,this paper proposes an automatic registration method for FFA image sequences based on mutual information,and applies the registered FFA image sequence to analyze the retinal blood flow velocity..The main contributions of the present work are as follows:(1)An automatic image registration method for FFA image sequences based on mutual information is proposed.In the method,the FFA image is first filtered by a multi-scale linear filtering method,the normalization principle is then used to increase the contrast between the blood vessel point and the background point,followed by threshold segmentation to obtain the rough blood vessel network in each frame of the FFA image;The pyramid downsample is used for the segmented image,then the mutual information is applied to calculate the similarity between the image to be registered and the reference image.The evolutionary strategy is used to optimize the registration parameters to obtain the spatial transformation matrix of the image with maximum mutual information,after which the FFA image sequence can be registered.(2)The retinal blood flow is analyzed based on the registered FFA image sequence.In this analysis,the region of interest in the reference image is first selected;according to the registered FFA image sequence,the time evolution of the gray intensity of each vascular point is extracted,and filtered by the median filter and the Gaussian filter.Then,the curve is fitted using 5-order polynomial.Feature parameters describing the retinal blood flow are extracted from the curve,such as starting time,peak time and perfusion time;finally,visualization method is used to generate a parameter image to reflect the blood flow velocity of the fundus.(3)In order to test the FFA image sequence registration algorithm proposed in this paper,four FFA angiography data(1039 frames)were collected from the local hospitals to construct a testing set.The registration accuracy,robustness and computational efficiency of the proposed registration algorithm are analyzed..The results show that the overall registration rate of the proposed algorithm reaches 93%,with the failure rate 1%.Compared with the commonly used registration methods(GDB-ICP,Glocker B),the proposed method has the advantages of high registration rate,fast speed and good robustness.It provides a good basis for further quantitative analysis of FFA image sequences.(4)FFA data of fourteen patients were collected from local hospitals,including 6 cases of central retinal vein occlusion,3 cases of diabetic fundus lesions,3 cases of paraventricular occlusion,and 2 cases of anterior ischemic lesions.Using the method proposed in this paper,the FFA image sequence of each patient is registered,based on which parametric image reflecting the retinal blood flow is generated.The parametric images of different patients were compared and analyzed.The results show that the parametric images can reflect the retinal blood flow velocity for different diseases,and is qualitatively consistent with the diagnosis of ophthalmologist.